I build systems for efficient AI inference.

Computer science student working across high-performance computing, inference engines, and the systems behind faster models.

Current focus

Sparse-first inference for long contexts.

Sparse-vLLM rethinks cache layout, controller flow, and kernels so sparse methods become first-class runtime choices.

Work shaped by profiling evidence, hardware constraints, and explicit performance goals.

Working across cache architecture, sparse runtime paths, benchmarks, and reproducible systems documentation.

Combined profiling, ONNX graph simplification, TensorRT FP16, persistent memory, and CUDA Graph capture.

Exposed parallelism, removed cache conflicts, and validated the optimized benchmark under the standard stability test.

May 2026 - Present

Research intern with the M³AIL Research Group at Harbin Institute of Technology, Shenzhen.

Oct 2025 - Present

Core member of the HITSZ supercomputing team, working on competition training and performance optimization.

Harbin Institute of Technology, Shenzhen

B.Eng. candidate, Computer Science and Technology2024 - Present

ASC Student Supercomputer Challenge

ASC26 Final, First Prize, 2026
Team captain

Huawei ICT Competition, China

National Final, Third Prize, 2025
Team captain

Cycling, hiking, travel, and places I want to remember.